Benchmarks¶

Three Python packages parse DPM-XL. pydpm-xl and dpmcore validate rules against a DPM database; dpm-xl evaluates rules against report data and explains failures. The table shows where they overlap and where they differ.

Capabilities¶

	dpm-xl	pydpm-xl	dpmcore
Licence	MIT	GPL-3.0-or-later	Apache-2.0
Runtime dependencies	none	8 (pandas, SQLAlchemy, database drivers)	antlr4, SQLAlchemy
Parses every EBA 3.4–4.4 rule	yes	yes	yes
Syntax errors	plain words with position	ANTLR messages	ANTLR messages
Evaluates rules against report data	yes	no	no
Explains failing records	yes	no	no
Columnar evaluation for millions of rows	yes	no	no
CLI to evaluate rules on a data file	yes	no	no
Tokenizer and editor grammar	yes	no	no
Outside dpm-xl's scope:
Semantic validation against a DPM release	no	yes	yes
DPM database migration, REST API	no	migration	yes
Reads report files	via dpm-xbrl	no	no

Use pydpm-xl or dpmcore to check that a rule fits a DPM release; use dpm-xl to check that a report satisfies the rules, and why not. The two combine.

Results¶

The corpus holds the 14,174 distinct rule and precondition expressions of EBA DPM releases 3.4 to 4.4. All three packages reject the same two, which contain typos in the EBA source.

Measured on 2026-10-24 on AMD Ryzen 7 4700U with Radeon Graphics, 8 cores, Linux 7.0.0-38-generic; Python 3.12.3. Best of 3 warm runs; peak RSS over those runs.

Speed and footprint¶

	dpm-xl 1.0.0	pydpm-xl 0.3.12	dpmcore 0.2.1
Syntax check of the corpus (whole process)	1.61 s, 23 MB	14.85 s, 109 MB	15.23 s, 113 MB
Per expression	113 µs	1,047 µs	1,075 µs
`import` (whole process)	0.09 s, 18 MB	0.75 s, 97 MB	0.85 s, 101 MB
Installed environment	1 distributions, 1 MB	17 distributions, 142 MB	15 distributions, 143 MB

Coverage of the EBA corpus¶

	dpm-xl 1.0.0	pydpm-xl 0.3.12	dpmcore 0.2.1
Expressions accepted	14,172 of 14,174	14,172 of 14,174	14,172 of 14,174
Rejected	`with {default: 0, interval: true}:if sum({tC_08.01.a, r0070…` `with {tV_03.01, default: null, interval: false}: if not(isn…`	`with {default: 0, interval: true}:if sum({tC_08.01.a, r0070…` `with {tV_03.01, default: null, interval: false}: if not(isn…`	`with {default: 0, interval: true}:if sum({tC_08.01.a, r0070…` `with {tV_03.01, default: null, interval: false}: if not(isn…`

DPM-XL constructs beyond the EBA rules¶

	dpm-xl 1.0.0	pydpm-xl 0.3.12	dpmcore 0.2.1
rank over a partition	accepted	rejected	accepted
windowed aggregate	accepted	rejected	accepted
statements with :=	accepted	rejected	rejected
persistent assignment <-	accepted	rejected	accepted
operation reference	accepted	accepted	accepted
set operators and literals	accepted	rejected	accepted
typed parameter	accepted	rejected	accepted
text concatenation &	accepted	accepted	accepted
date functions	accepted	rejected	accepted
date() and date literals	accepted	rejected	accepted
table group	accepted	accepted	accepted
annualise	accepted	rejected	accepted

Syntax error messages¶

{tC_01.00, r0010, c0010} >=

dpm-xl: expression ends where an operand is expected at position 27
pydpm-xl: offendingSymbol: [@11,27:26='',<-1>,1:27] msg: mismatched input '' expecting {BOOLEAN_LITERAL, NOT,…
dpmcore: Unexpected error: "offendingSymbol: [@11,27:26='',<-1>,1:27] msg: mismatched input '' expecting {BO…

with {tC_01.00, c0010}: {r0010} = {r0020} +

dpm-xl: expression ends where an operand is expected at position 43
pydpm-xl: offendingSymbol: [@21,43:42='',<-1>,1:43] msg: mismatched input '' expecting {BOOLEAN_LITERAL, NOT,…
dpmcore: Unexpected error: "offendingSymbol: [@21,43:42='',<-1>,1:43] msg: mismatched input '' expecting {BO…

Method¶

Each package runs in its own virtual environment from PyPI (dpm-xl from this repository), as a separate process through its public syntax-check API: one warm-up, then the best of three runs, with peak memory. Ratios between the packages are more stable than absolute times. Scripts and instructions are in benchmarks/.